Create api

.gitignore

@@ -8,3 +8,6 @@ __pycache__
 magnolia.egg-info/
 dist/
 build/
+magnolia/python/api/model/
+magnolia/python/api/converted/
+magnolia/python/api/uploads/

magnolia/data/logging_settings/logging.conf

@@ -1,8 +1,8 @@
 [loggers]
-keys=root,preprocessing,partitioning,model
+keys=root,preprocessing,partitioning,model,api
 
 [handlers]
-keys=console_handler,verbose_console_handler,preprocessing_file_handler,partitioning_file_handler,model_file_handler
+keys=console_handler,verbose_console_handler,preprocessing_file_handler,partitioning_file_handler,model_file_handler, api_file_handler
 
 [formatters]
 keys=brief_formatter,verbose_formatter
@@ -30,6 +30,12 @@ handlers=verbose_console_handler,model_file_handler
 qualname=model
 propagate=1
 
+[logger_api]
+level=INFO
+handlers=verbose_console_handler,api_file_handler
+qualname=api
+propagate=1
+
 
 [handler_console_handler]
 class=StreamHandler
@@ -61,6 +67,12 @@ level=DEBUG
 formatter=verbose_formatter
 args=('model.log', 'w', 10000000, 5)
 
+[handler_api_file_handler]
+class=handlers.RotatingFileHandler
+level=DEBUG
+formatter=verbose_formatter
+args=('api.log', 'w', 10000000, 5)
+
 
 [formatter_brief_formatter]
 format=%(levelname)s | %(funcName)s | %(asctime)s | %(message)s

magnolia/python/api/api.py

@@ -0,0 +1,140 @@
+import json
+import os
+import tempfile
+import logging.config
+from flask import Flask, request, jsonify, url_for, send_from_directory
+from flask_cors import CORS, cross_origin
+import librosa
+import numpy as np
+
+from magnolia.preprocessing.preprocessing import normalize_waveform
+from magnolia.preprocessing.preprocessing import preprocess_waveform
+from magnolia.preprocessing.preprocessing import undo_preprocessing
+from magnolia.utils.postprocessing import convert_preprocessing_parameters
+from magnolia.utils.clustering_utils import chimera_clustering_separate, chimera_mask
+from magnolia.models import make_model
+
+UPLOAD_DIR = './uploads'
+CONVERTED_DIR = './converted'
+MODEL_DIR = './model'
+MODEL_DEVICE = '/cpu:0'
+ALLOWED_EXTENSIONS = set(['wav'])
+SAMPLING_RATE = 16000
+
+settings={
+    "compression": "gzip",
+    "compression_opts": 0,
+    "processing_parameters": {
+        "target_sample_rate": 10000,
+        "preemphasis_coeff": 0.0,
+        "stft_args": {
+            "n_fft": 512,
+            'hop_length': 256
+        }
+    }
+}
+
+app = Flask(__name__)
+CORS(app)
+app.config['UPLOAD_DIR'] = UPLOAD_DIR
+app.config['CONVERTED_DIR'] = CONVERTED_DIR
+app.config['MODEL_DIR'] = MODEL_DIR
+
+def valid_file(filename):
+    return '.' in filename and \
+           filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS
+
+def save_temp_file(request):
+    if 'file' not in request.files:
+        flash('No file part')
+        return ''
+    file = request.files['file']
+    # if user does not select file, browser also
+    # submit a empty part without filename
+    if file.filename == '':
+        flash('No selected file')
+        return ''
+    if file and valid_file(file.filename):
+        tf = tempfile.NamedTemporaryFile()
+        filename = os.path.basename(tf.name)
+        temp_path = os.path.join(app.config['UPLOAD_DIR'], filename)
+        file.save(temp_path)
+        return temp_path
+
+
+def convert_file_to_waveform(file_path):
+    y,sr = librosa.load(file_path)
+    print("sampling rate: {0}".format(sr))
+    duration = librosa.get_duration(y=y, sr=sr)
+    print("durations: {0}".format(duration))
+    print("len in bits: {0}".format(int(duration*sr)))
+    return y, sr, duration
+
+def prepare_waveform(wf_batch):
+    # Compute STFT spectrogram
+    signal = wf_batch #normalize_waveform(np.squeeze(wf_batch))
+    D,yy = preprocess_waveform(wf_batch, SAMPLING_RATE, **settings['processing_parameters'])
+    app.logger.debug(D.shape)
+    app.logger.debug(yy.shape)
+    return D
+
+def process_waveform(D):
+    frequency_dim = D.shape[0]
+    print("D shape 1: {0}".format(D.shape[1]))
+    model_save_base = MODEL_DIR
+    model_location = MODEL_DEVICE
+
+    model_params = {
+            'layer_size': 500,
+            'embedding_size': 10,
+            'alpha': 0.1,
+            'nonlinearity': 'tf.tanh',
+        }
+    model_params['F'] = frequency_dim
+    config = {'model_params': model_params,
+              'device': model_location}
+    model = make_model('Chimera', config)
+    model.load(model_save_base)
+
+    source_specs = chimera_mask(np.expand_dims(D,axis=0), model)[0]
+    print(source_specs[0].shape)
+    return source_specs
+
+def postprocess_waveform(source_specs, sample_length_in_bits):
+    istft_args={'hop_length': 256, 'win_length': 512}
+    yy_out = undo_preprocessing(source_specs[:,:,0], sample_length_in_bits,
+                                   preemphasis_coeff=0,
+                                   istft_args=istft_args)
+
+    return yy_out
+
+def convert_back_to_wav(yy, save_sr):
+    tf = tempfile.NamedTemporaryFile()
+    filename = os.path.basename(tf.name)
+    temp_path = os.path.join(app.config['CONVERTED_DIR'], filename)
+    librosa.output.write_wav(temp_path, yy, save_sr)
+    return url_for('uploaded_file', filename=filename)
+
+@app.route('/api/v1/converted/<filename>', methods=['GET'])
+@cross_origin(origin='*')
+def uploaded_file(filename):
+    return send_from_directory(app.config['CONVERTED_DIR'],
+                               filename, attachment_filename=filename + '.wav', mimetype='audio/wav')
+
+@app.route('/api/v1/convert', methods=['POST'])
+@cross_origin(origin='*')
+def convert_file():
+    temp_path = save_temp_file(request)
+    wf_batch, load_sr, len_in_secs = convert_file_to_waveform(temp_path)
+    SAMPLING_RATE = load_sr
+    len_in_bits = int(16000*len_in_secs)
+    D = prepare_waveform(wf_batch)
+    source_specs = process_waveform(D)
+    yy = postprocess_waveform(source_specs, len_in_bits)
+    converted_url = convert_back_to_wav(yy, 13000)
+    return converted_url
+
+if __name__ == '__main__':
+    app.logger = logging.getLogger('api')
+    app.logger.debug("logging started")
+    app.run(port='5001', debug=True)

magnolia/python/utils/clustering_utils.py

@@ -9,8 +9,8 @@
 from sklearn.mixture import BayesianGaussianMixture, GaussianMixture
 from sklearn.decomposition import PCA
 
-from ..features.spectral_features import istft
-from ..features.preprocessing import make_stft_features, \
+from magnolia.preprocessing.spectral_features import istft
+from magnolia.preprocessing.preprocessing import make_stft_features, \
                                      undo_preemphasis
 from magnolia.utils.training import preprocess_l41_batch, preprocess_chimera_batch, preprocess_l41_regression_batch
 

magnolia/python/utils/postprocessing.py

@@ -1,6 +1,6 @@
 import numpy as np
-from ..features.preprocessing import undo_preemphasis
-from ..features.spectral_features import istft
+from magnolia.preprocessing.preprocessing import undo_preemphasis
+from magnolia.preprocessing.spectral_features import istft
 
 
 def convert_preprocessing_parameters(params):